c++ - Object Detection: Training Requried or No Training Required? -


This question is related to object detection, and basically, detecting any "known" objects. For example, I have imagined things below.

  1. Table
  2. Bottle
  3. The camera
  4. car

I will take 4 photos from all these personal objects. From one to the left, from the second to the right, and the second 2 up and down I originally thought that it is possible to identify these objects with these 4 photos per each, because you have photos in all 4 angles, whatever the object Find out if you do not see him

But I got confused, there is no idea about training the engine with thousands positive and negative images from each object. I do not really think this is necessary.

So just say, my question is, to identify an object, do I need these thousands of positive and negative objects? Or just 4 photos enough from 4 angle?

I am hoping to use OpenCV for this.

Update

In fact, the main thing is something like this .. Imagine that I have 2 laptops and one is Dell But you know that the logo is clearly visible. Can we use it? If not, how to "training" the "hard work" process? How many pictures are needed?

Update 2 I need to find "specific" items not all cars, all bottles, etc. For example, "Maruti car model 123" and "Ferrari car model 234" are both cars but are different. Imagine my above model has photos of Maruti and Ferrari, so I have to find out. I do not have to worry about other cars or vehicles or other models of Maruti and Ferrari. But the above mentioned "Maruti car model 123" should be identified as "Maruti car model 123" and the above mentioned "Ferrari car model 234" should be identified as "Ferrari car model 234".

  1. If you find a specific object address And you do not need an account for point-of-change, you can use the 2D feature:

  2. To differentiate between 2 logos, you can Probably each logo will need to create a detector that will be trained on the set of images. For example, you can train Higher Cascade Classifier.

  3. To distinguish between different models of cars, you will probably need to train the classifier using the training images of each car. However, I came to an app that is using the closest neighbor approach - it only removes the characteristics from the given test images and compares it to the address of the images of the difference car model.

In addition, I can recommend some methods and packages if you explain more on the application.


Comments

Popular posts from this blog

Pass DB Connection parameters to a Kettle a.k.a PDI table Input step dynamically from Excel -

multithreading - PhantomJS-Node in a for Loop -

c++ - MATLAB .m file to .mex file using Matlab Compiler -